https://www.duolingo.com/mercutio

How do weird duolingo sentences get generated?

Duolingo sometimes seems to have hilarious and odd sentences that occur in it's lessons and there's even a twitter page dedicated to it. I've been curious HOW these sentences occur? As presumably they are not programmed in..... Such things as "I'm a penguin" and "the waitress has nine fingers" Two that personally occurred on mine have been " the woman is in my bedroom" and "my father is the son of peasants"..... Can anyone explain?

4 years ago

17 Comments


https://www.duolingo.com/flandybandy

I have no insider knowledge on this, but my best guess as to how Duolingo works is the following:

  • All sentences are in fact manually generated!
  • Each language have a long list of sentences Duolingo can use.
  • Each sentence has a list of the vocabulary it uses as well as each grammar rules it uses.
  • Each sentence also has a list of accepted translations.
  • Each lesson have a list with the grammar/vocabulary that it is supposed to teach.
  • At each step in a lesson a sentence is chosen from the list of of sentences of the language such that:
  • It uses a word or a grammar rule that the lesson is supposed to teach.
  • It uses no word or grammar rules that is not already known or supposed to taught in the lesson.

My reasons for believing this are the following:

  • The number of sentences seems quite limited, which would not be expected if they were randomly generated.
  • Duolingo seems quite good at accepting solutions that maintain the meaning of the sentence, but are far from a direct translation. If the sentences were randomly generated, so would the translations have to be.
  • One translation I saw had an error that clearly indicated that the translation was made manually. (A space in the middle of a word)
  • When you revisit an old lesson Duolingo includes some sentences that uses words and grammar rules that are only taught further down the language tree.
  • The sentences are to some degree meaningful eg. if you lost a finger you would have nine, the colours of objects are usually correct, etc..

I might be mistaken, but I cannot think of any other explanation that explains all my observations.

4 years ago

https://www.duolingo.com/stecchetto
stecchettoPlus
  • 25
  • 20
  • 18
  • 17
  • 14
  • 14
  • 13
  • 10
  • 10
  • 9
  • 9
  • 9
  • 8
  • 6
  • 6
  • 6
  • 6
  • 6
  • 5
  • 5
  • 5
  • 4
  • 4
  • 4
  • 2
  • 2

This sounds about right. The peek at the Incubator interface rkuprov posted a while ago gives us some clues and seems mostly consistent with what you're saying:

http://www.duolingo.com/comment/1007394

4 years ago

https://www.duolingo.com/Luscinda
Luscinda
  • 25
  • 25
  • 25
  • 15
  • 11
  • 10
  • 8
  • 7
  • 6
  • 6
  • 5
  • 4
  • 3
  • 3
  • 3
  • 2

The real giveaway is the recordings. Someone is writing this stuff, not randomly generating it. It's all deliberate.

4 years ago

https://www.duolingo.com/LingPenguin
LingPenguin
  • 17
  • 15
  • 14
  • 12
  • 11
  • 11
  • 97

I think that Duolingo has some sort of an "adlibs" type system, where they have a sentence and just say that this word should be a noun and this one should be an article of clothing and this one should be food, and then it takes them and puts in words that you're learning or need to work on. With that said though, I have had no part in developing Duolingo and have not heard this from anybody who does, this is just my theory.

Also, my favourite weird sentence is near the end of the Spanish tree, and it roughly translates to: "She used to carry the tomato from America".

4 years ago

https://www.duolingo.com/mercutio

I've wondered for a while if the wierd sentences are the same for everyone or whether the wierd sentences are somehow self generated by accident and therefore each user gets different wierd sentences

4 years ago

https://www.duolingo.com/ProwlingParis
ProwlingParis
  • 14
  • 12
  • 10
  • 10
  • 9

I mod ENG-for-TR and personally can't resist translating sentences like "the penguin is green today" when I see them. So visceral and all, I imagine it leaves a unique image in the learner's mind and has a further lasting memory.

Also there are anywhere between a handful and a buttload of sentences to choose from for each word, of which I need to translate at least 3. With the many options it's usually possible to find the most natural sounding sentences for the language pair that is being taught, but sometimes it's simply not good enough. And how those sentences are generated in the first place is that I imagine it's a randomly process at first, but are filtered for nonsensical ones later on--would me the most efficient operation imo. Also you can assume that all of those sentences make sense in one language or another as they are, but get wacky once translated; very tricky :-)

4 years ago

https://www.duolingo.com/chilvence
chilvence
  • 14
  • 13
  • 12
  • 12
  • 12
  • 12
  • 12
  • 12
  • 12
  • 12
  • 12
  • 12
  • 12
  • 12
  • 12
  • 12
  • 10
  • 6
  • 6
  • 5
  • 4
  • 2
  • 2

In honour of this quirkiness, my Duolingo avatar is now a duck that contains fruit. I invite everyone else to do the same thing and artistically represent their favourite phrase on their profile pic.

4 years ago

https://www.duolingo.com/lilygilder
lilygilder
  • 11
  • 11
  • 11
  • 10
  • 7

That's very creative of you! Cool!

4 years ago

https://www.duolingo.com/hdcanis
hdcanis
  • 22
  • 17
  • 3

And sometimes you get clear pop culture references ("So long and thanks for all the fish") but there might be some adlib system where the sentences are mostly written out manually but with some gaps allowing variation. Those "which of these three sentences are correct" questions would at least suggest that...

One wonders though about those "creepy stalker" sentences, the "I hide the child in my cellar" type.

4 years ago

https://www.duolingo.com/SD-77
SD-77
  • 22
  • 18
  • 16
  • 15
  • 15
  • 15
  • 14
  • 11
  • 9
  • 9
  • 9
  • 9
  • 8
  • 7
  • 6

Or "The girl walks in the shadows." Or "You can die."

4 years ago

https://www.duolingo.com/HerrArbo
HerrArbo
  • 20
  • 10
  • 9
  • 9
  • 6
  • 6
  • 6
  • 3
  • 2

Strange things stand out which makes them more memorable. So I guess this is on purpose to make sure we remember these things and the extra vocab it opens up to us. Also, forigners are a funny bunch, so phrases like "the yellow cow reads the newspaper to the gentlemen" is probably quite common amongst, say, the French or the Dutch in compason to us English speaking folk.

4 years ago

https://www.duolingo.com/Susande
Susande
  • 21
  • 13
  • 12
  • 11
  • 140

You're right, but I think the French are more used to yellow roosters reading their newspapers.

4 years ago

https://www.duolingo.com/mikevir
mikevir
  • 25
  • 25
  • 22
  • 19
  • 17
  • 10
  • 184

I for one don't think these sentences are that weird. I would explain further, but my penguin, who is drinking milk, wants some rice.

4 years ago

https://www.duolingo.com/adamyoung97
adamyoung97
  • 15
  • 15
  • 15
  • 14
  • 13
  • 13
  • 12
  • 10
  • 9
  • 8
  • 8

"The horse eats the girl"

4 years ago

https://www.duolingo.com/BoulderSpanish

We have to remember that people from many cultures are learning these languages. My parents were not peasants but some people have parents who are peasants. There are also overtones to the use of certain words that we have to consider. Something to think about.

4 years ago

https://www.duolingo.com/Diggabyte
Diggabyte
  • 22
  • 12
  • 8
  • 6
  • 6
  • 5
  • 4
  • 3

I thought that in order to pay for hosting, duolingo was having it's users translate sentences from the web, so maybe these are all sentences from the websites that duolingo gets payed to translate? I'm not entirely sure on how there system of user translations works though.

4 years ago

https://www.duolingo.com/dinkiedoo42

I got "the cat eats a book"

4 years ago
Learn a language in just 5 minutes a day. For free.

Learn a language in just 5 minutes a day. For free.